diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 419268ee985aba62539dd16fe865574fde67ca32..2623a188b5d5ce345de344fa21f0daf636486762 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -177,8 +177,6 @@ variables:
 .linux-64:
   extends: .linux
   variables:
-    # set fake cuda version for __cuda virtual package
-    CONDA_OVERRIDE_CUDA: "11.2"
     SUBDIR: "linux-64"
 
 .linux-aarch64:
@@ -598,7 +596,6 @@ test:docker:
           --tag "${DOCKER_TAG}"
           --build-arg "INSTALL_PATH=${CVMFS_PATH}"
           --build-arg "BUILD_OPTS=${BUILD_OPTS}"
-          --build-arg "CONDA_OVERRIDE_CUDA=${CONDA_OVERRIDE_CUDA}"
     - docker push "${DOCKER_TAG}"
   retry: 2
 
@@ -627,8 +624,6 @@ build:staging:
 build:testing:
   extends:
     - .build
-  variables:
-    CONDA_OVERRIDE_CUDA: "11.2"
   rules:
     # only for production testing builds
     - *if-testing
diff --git a/packages/dingo-gw.yaml b/packages/dingo-gw.yaml
index 19a636d1a32c6e8cfdf5d77b5536c95095d1828c..28c3529eec6c818be01e1b1e0107767c4e953dd8 100644
--- a/packages/dingo-gw.yaml
+++ b/packages/dingo-gw.yaml
@@ -12,9 +12,5 @@ skip: true  # [linux and not x86_64]
 # needs pyseobnr
 skip: true  # [osx and not x86_64]
 
-dependencies:
-  # use the cuda build of pytorch on Linux x86_64
-  - pytorch=*=cuda*  # [linux64]
-
 test:
   script: dingo-gw-tests.sh -k 'not EOB and not FD_f_max_failure'
diff --git a/scripts/build-cvmfs.dockerfile b/scripts/build-cvmfs.dockerfile
index 3cd76458a713c0916dc847443a963cd58d42c021..c530454b4972ea58511846c63984771f67f79442 100644
--- a/scripts/build-cvmfs.dockerfile
+++ b/scripts/build-cvmfs.dockerfile
@@ -25,11 +25,9 @@ FROM debian:buster AS builder
 ARG INSTALL_PATH=/cvmfs/software.igwn.org/conda/
 ARG SUBDIR=linux-64
 ARG BUILD_OPTS=""
-ARG CONDA_OVERRIDE_CUDA="11.2"
 
 ENV LANG=C.UTF-8 LC_ALL=C.UTF-8
 ENV PATH=${INSTALL_PATH}/condabin:${PATH}
-ENV CONDA_OVERRIDE_CUDA=${CONDA_OVERRIDE_CUDA}
 
 RUN apt-get update --quiet --fix-missing && \
     apt-get install --quiet --quiet --yes \